File size: 4,706 Bytes
f24de5e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
99e1119
 
 
 
 
 
 
f24de5e
 
99e1119
 
f24de5e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
99e1119
 
f24de5e
 
 
 
 
 
 
 
 
f1f3abc
 
 
f24de5e
 
 
 
 
 
 
 
3207007
f24de5e
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
---
license: mit
---
# intel-optimized-model-for-embeddings-v1

This is a text embedding model model: It maps sentences & paragraphs to a 512 dimensional dense vector space and can be used for tasks like clustering or semantic search. For sample code that uses this model in a torch serve container see [Intel-Optimized-Container-for-Embeddings](https://github.com/intel/Intel-Optimized-Container-for-Embeddings).

## Usage

Install the required packages:
```
pip install -U torch==2.3.1+cpu --extra-index-url https://download.pytorch.org/whl/cpu
pip install -U transformers==4.42.4 intel-extension-for-pytorch==2.3.100
```

Use the following example below to load the model with the transformers library, tokenize the text, run the model, and apply pooling to the output.

```
import torch
from transformers import AutoTokenizer, AutoModel
import intel_extension_for_pytorch as ipex

def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0]
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded,
                        1) / torch.clamp(input_mask_expanded.sum(1),
                                        min=1e-9)

# load model
tokenizer = AutoTokenizer.from_pretrained('Intel/intel-optimized-model-for-embeddings-v1')
model = AutoModel.from_pretrained('Intel/intel-optimized-model-for-embeddings-v1', 
                                   torchscript=True)
model.eval()

# do IPEX optimization
batch_size = 1
seq_length=512
vocab_size = model.config.vocab_size
sample_input = {"input_ids": torch.randint(vocab_size, size=[batch_size, seq_length]),
                "token_type_ids": torch.zeros(size=[batch_size, seq_length],
                                                dtype=torch.int),
                "attention_mask": torch.randint(1, size=[batch_size, seq_length])}
text = "This is a test."
model = ipex.optimize(model, level="O1",auto_kernel_selection=True,
                            conv_bn_folding=False, dtype=torch.bfloat16)

with torch.no_grad(), torch.cpu.amp.autocast(cache_enabled=False,
                                                dtype=torch.bfloat16):
    # Compile model
    model = torch.jit.trace(model, example_kwarg_inputs=sample_input,
                                    check_trace=False, strict=False)
    model = torch.jit.freeze(model)
    
    # Call model
    tokenized_text = tokenizer(text, padding=True, truncation=True, return_tensors='pt')
    model_output = model(**tokenized_text)
    sentence_embeddings = mean_pooling(model_output,tokenized_text['attention_mask'])
    embeddings = sentence_embeddings[0].tolist()

# Embeddings output
print(embeddings)
```

## Model Details

### Model Description

This model was fine-tuned using the [sentence-transformers](https://github.com/UKPLab/sentence-transformers) library 
based on the [BERT-Medium_L-8_H-512_A-8](https://huggingface.co/nreimers/BERT-Medium_L-8_H-512_A-8) model
 using [UAE-Large-V1](https://huggingface.co/WhereIsAI/UAE-Large-V1) as a teacher.


### Training Datasets

| Dataset       | Description           | License  |
| ------------- |:-------------:| -----:|
| beir/dbpedia-entity      | DBpedia-Entity is a standard test collection for entity search over the DBpedia knowledge base.  | CC BY-SA 3.0 license |
| beir/nq      | To help spur development in open-domain question answering, the Natural Questions (NQ) corpus has been created, along with a challenge website based on this data.       |   CC BY-SA 3.0 license |
| beir/scidocs | SciDocs is a new evaluation benchmark consisting of seven document-level tasks ranging from citation prediction, to document classification and recommendation.       |    CC-BY-SA-4.0 license  |
| beir/trec-covid | TREC-COVID followed the TREC model for building IR test collections through community evaluations of search systems.       |  CC-BY-SA-4.0 license  |
| beir/touche2020 | Given a question on a controversial topic, retrieve relevant arguments from a focused crawl of online debate portals.      |    CC BY 4.0 license  |
| WikiAnswers | The WikiAnswers corpus contains clusters of questions tagged by WikiAnswers users as paraphrases.       |    MIT |
| Cohere/wikipedia-22-12-en-embeddings Dataset  | The Cohere/Wikipedia dataset is a processed version of the wikipedia-22-12 dataset. It is English only, and the articles are broken up into paragraphs.       |    Apache 2.0  |
| MLNI  | GLUE, the General Language Understanding Evaluation benchmark (https://gluebenchmark.com/) is a collection of resources for training, evaluating, and analyzing natural language understanding systems.       |    MIT |